OpenAI Releases Full o1 Reasoning Model to Select Developers, Enhancing AI Capabilities and Pricing

OpenAI Unveils Full o1 Reasoning Model to Select Developers

OpenAI has announced the release of its full o1 reasoning model to select developers through its API, marking a significant advancement in AI technology. This release is part of OpenAI's "12 Days of OpenAI" holiday-themed press blitz, occurring on the ninth day of the event 1

1

2

2

.

Access and Pricing

The full o1 model is initially available to developers in OpenAI's "Tier 5" category, which includes users who have had an account for over a month and spend at least $1,000 monthly on OpenAI services 1

1

3

3

. The pricing for o1 is notably higher than previous models, reflecting the increased computing resources required:

• $15 for analyzing approximately 750,000 words
• $60 for generating approximately 750,000 words

This pricing structure makes o1 three to four times more expensive than the GPT-4o model 2

2

4

4

.

Enhanced Capabilities

The full o1 model offers several improvements over its preview version:

1. Fact-checking: o1 can effectively fact-check itself, reducing errors and hallucinations 4

   4

   .
2. Customization: A new "reasoning_effort" parameter allows control over the model's processing time 1

   1

   3

   3

   .
3. Function calling: Enables connection to external data sources 1

   1

   .
4. Developer messages: Allows instruction on tone and style 3

   3

   .
5. Image analysis: Adds visual reasoning capabilities 1

   1

   3

   3

   .
6. Improved efficiency: Uses 60% fewer reasoning tokens on average 1

   1

   .

Realtime API Updates

OpenAI has also announced updates to its Realtime API, designed for low-latency vocal AI applications:

1. Integration of GPT-4o and 4o-mini models 1

   1

   2

   2

   .
2. Support for WebRTC, facilitating smooth audio communication in web browsers 2

   2

   4

   4

   .
3. Reduced costs for audio tokens: 60% reduction for o1 and 90% for mini tokens 4

   4

   .

New Fine-tuning Method

OpenAI introduced a new fine-tuning method called "direct preference optimization" or "preference fine-tuning":

• Allows model trainers to provide two outputs and specify a preference 4

  4

  .
• Enhances the model's ability to adapt to various response styles, formatting, and helpfulness 4

  4

  5

  5

  .

Developer Tools and Future Plans

1. New software development kits (SDKs) for Go and Java programming languages 4

   4

   5

   5

   .
2. Plans to expand access and increase rate limits for more developers beyond the initial tier 5 category 4

   4

   .

Industry Impact

The release of the full o1 model represents a significant step in AI development, offering more advanced reasoning capabilities to developers. However, the high cost may limit its accessibility, potentially impacting its adoption rate and use cases in various industries 1

1

2

2

3

3

4

4

5

5

.